Automated Hybrid Propensity Decision Vector Generation Using Artificial Intelligence

1. A computer implemented method to automatically generate an optimized decision distribution vector for a plurality of related, demand-correlated products, comprising: receiving, by said computer, a data set having a plurality of entries indexed according to said plurality of related, demand-correlated products, with each entry including a plurality of entry attributes; receiving, by said computer, current decision context data for said plurality of products, said current decision context data selected from a list consisting of price data, inventory data, promotion campaign data, and advertisement data; determining, by said computer, a set of primary entry attributes from among said plurality of entry attributes; training, by said computer, a first machine learning model based upon said set of primary entry attributes offline or during non-peak times; responsive to said computer receiving a decision request including a set of attributes that includes an associated set of attributes corresponding to said set of primary entry attributes, scoring by said computer, said associated set of attributes corresponding to said set of primary entry attributes with said first machine learning model to generate, by said computer, a baseline purchase propensity; to determine incrementally modified price elasticities for the plurality of entry attributes using the baseline purchase propensity; training, by said computer, a second machine learning model, based incrementally upon each of said entry attributes to generate, by said computer, incrementally modified own-product elasticity data and cross-product elasticity data associated therewith; determining, by said computer, a pricing effectiveness value associated with each of said entry attributes for each of said associated incrementally modified price elasticities; ranking, by said computer, said entry attributes by said calculated pricing effectiveness value; iteratively adding, by said computer, each of said entry attributes in decreasing pricing effectiveness rank order to a set of secondary entry attributes used with said second machine learning model to calculate said decision distribution vector, until an own-product elasticity data and cross-product elasticity data computation duration exceeds a selected computation duration threshold, labeling, by said computer, each remaining entry attribute as a primary entry attribute and removing each of said entry attributes added to said secondary entry attributes from said set of primary entry attributes; responsive to generating said baseline purchase propensity, training by said computer, the second machine learning model based upon said baseline purchase propensity and said current decision context data, to generate own- and cross-product elasticity data with respect to the current decision context data, the baseline propensity being combined with the current decision context data and the secondary event attributes to generate the own-and cross-product elasticity data which indicates a customer sensitivity to prices, the current decision context data including pricing assigned to various products purchased; generating, by said computer, using said own- and cross-product elasticity data, a decision distribution vector selected, at least in part, in accordance with said current decision context data for said plurality of related, demand-correlated products, the distribution vector including prices for each of the options a customer chooses to meet travel criteria included in the decision request; providing, by said computer via a customer display menu in operative communication with said computer, a set of purchase choices representing a price distribution based, at least in part, on the distribution vector; responsive to providing the set of purchase choices, receiving by the computer, an indication of user selection; and responsive to receiving the indication of user selection, adding metadata associated with the user selection for use in updated training of the first and second machine learning models.

2. The computer implemented method of claim 1 , wherein said pricing effectiveness is based upon considerations selected from a list consisting of predicted revenue error and revenue lift.

3. The computer implemented method of claim 1 , wherein said selected computation duration threshold is 10 milliseconds.

4. The computer implemented method of claim 1 , wherein said decision context data further includes attributes selected from a list consisting of available inventory and origin-destination data.

5. The computer implemented method of claim 1 , wherein said entry attributes are selected from a list consisting of customer contexts, including demographics, loyalty, frequency/recency, and preferences; purchase contexts, including time of day, day of week, booking group size, advance purchase aspects; market contexts, including location features, market type, destination; and product contexts, including quality aspects, stayover restrictions.

6. The computer implemented method of claim 1 , wherein said first machine learning model is a recurrent neural network.

7. A system to automatically generate an optimized decision distribution vector for a plurality of related, demand-correlated products, which comprises: a computer system comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by the computer to cause the computer to: receive a data set having a plurality of entries indexed according to said plurality of related, demand-correlated products, with each entry including a plurality of entry attributes; receive current decision context data for said plurality of products, said decision current context data selected from a list consisting of price data, inventory data, promotion campaign data, and advertisement data; determine a set of primary entry attributes from among said plurality of entry attributes; train a first machine learning model based upon said set of primary entry attributes offline or during non-peak times; responsive to said computer receiving a decision request including a set of attributes that includes an associated set of attributes corresponding to said set of primary entry attributes, score said associated set of attributes corresponding to said set of primary entry attributes with said first machine learning model to generate a baseline purchase propensity; determine incrementally modified price elasticities for the plurality of entry attributes using the baseline purchase propensity; train a second machine learning model, based incrementally upon each of said entry attributes to generate incrementally modified own-product and cross-product elasticity data associated therewith; determine a pricing effectiveness value associated with each of said entry attributes for each of said associated incrementally modified price elasticities; rank said entry attributes by said calculated pricing effectiveness value; iteratively add each of said entry attributes in decreasing pricing effectiveness rank order to a set of secondary attributes used with said second machine learning model to calculate said decision distribution vector, until an own-product and cross-product elasticity data computation duration exceeds a selected computation duration threshold, and label each remaining entry attribute as a primary entry attribute and remove each of said entry attributes added to said secondary entry attributes from said set of primary attributes; responsive to generating said baseline purchase propensity, train the second machine learning model based upon said baseline purchase propensity and said current decision context data, to generate own-product and cross-product elasticity data, the baseline propensity is combined with the current decision context data and the secondary event attributes to generate price elasticity, which indicates a customer sensitivity to prices, the current decision context data including pricing assigned to various products purchased; generate, using said own-product elasticity data and cross-product elasticity data, a decision distribution vector selected, at least in part, in accordance with said current decision context data for said plurality of related, demand-correlated products, the distribution vector including prices for each of the options a customer chooses to meet travel criteria included in the decision request; provide via a customer display menu in operative communication with said computer, a set of purchase choices representing a price distribution based, at least in part, on the distribution vector; responsive to providing the set of purchase choices, receive, an indication of user selection and updating the data set with attributes of the user selection; and responsive to receiving the indication of user selection, add metadata associated with the user selection for use in updated training of the first and second machine learning models.

8. The system of claim 7 , wherein said pricing effectiveness is based upon considerations selected from a list consisting of predicted revenue error and revenue lift.

9. The system of claim 7 , wherein said selected computation duration threshold is 10 milliseconds.

10. The system of claim 7 , wherein said decision context data further includes attributes selected from a list consisting of available inventory and origin-destination data.

11. The system of claim 7 , wherein said entry attributes are selected from a list consisting of customer contexts, including demographics, loyalty, frequency/recency, and preferences; purchase contexts, including time of day, day of week, booking group size, advance purchase aspects; market contexts, including location features, market type, destination; and product contexts, including quality aspects, stayover restrictions.

12. The system of claim 7 , wherein said first machine learning model is a recurrent neural network.

13. A computer program product to automatically generate an optimized decision distribution vector for a plurality of related, demand-correlated products, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by the computer to cause the computer to: receive, using the computer, a data set having a plurality of entries indexed according to said plurality of related, demand-correlated products, with each entry including a plurality of entry attributes; receive, using the computer, current decision context data for said plurality of products, said current decision context data selected from a list consisting of price data, inventory data, promotion campaign data, and advertisement data; determine, using the computer, a set of primary entry attributes from among said plurality of entry attributes; train, using the computer, a first machine learning model based upon said set of primary entry attributes offline or during non-peak times; responsive to said computer receiving a decision request including a set of attributes that includes an associated set of attributes corresponding to said set of primary entry attributes, score using the computer, said associated set of attributes corresponding to said set of primary entry attributes with said first machine learning model to generate a baseline purchase propensity; determine incrementally modified price elasticities for the plurality of entry attributes using the baseline purchase propensity; train, using the computer, said second machine learning model, based incrementally upon each of said entry attributes to generate incrementally modified own-product elasticity data and cross-product elasticity data associated therewith; determine, using the computer, a pricing effectiveness value associated with each of said entry attributes for each of said associated incrementally modified price elasticities; rank, using the computer, said entry attributes by said calculated pricing effectiveness value; iteratively add, using said computer, each of said entry attributes in decreasing pricing effectiveness rank order to a set of secondary attributes used with said second machine learning model to calculate said decision distribution vector, until an own-product elasticity data and cross-product elasticity data computation duration exceeds a selected computation duration threshold, and label, using said computer, each remaining entry attribute as a primary entry attribute and remove, using the computer, each of said entry attributes added to said secondary entry attributes from said set of primary attributes; responsive to generating said baseline purchase propensity, train, using the computer, a second machine learning model based upon said baseline purchase propensity and said current decision context data, to generate own-product elasticity data and cross-product elasticity data, with respect to the current decision context data, the baseline propensity is combined with the current decision context data and the secondary event attributes to generate price elasticity, which indicates a customer sensitivity to prices, the decision context data including pricing assigned to various products purchased; generate, using said computer and said own-product elasticity data and cross-product elasticity data, a decision distribution vector selected, at least in part, in accordance with said current decision context data for said plurality of related, demand-correlated products, the distribution vector including prices for each of the options a customer chooses to meet travel criteria included in the decision request; provide via a customer display menu in operative communication with said computer, a set of purchase choices representing a price distribution based, at least in part, on the distribution vector; responsive to providing the set of purchase choices, receive, an indication of user selection and updating the data set with attributes of the user selection; and responsive to receiving the indication of user selection, add metadata associated with the user selection for use in updated training of the first and second machine learning models.

14. The computer program product of claim 13 , wherein said first machine learning model is a recurrent neural network.